Post on 17-Apr-2018
Thanks to
� GL. Caporello and A. Maravall (and the Bank of Spain)
� Documentation, support, code, ...
� R&D Unit of the Department of Statistics of the NBB.� F. Osaer (DBs, statistical tools, X11)
� Y. Molitor (Graphical user interfaces)
� P. Hianne (Excel Add-ins)
Plan
� Context� Technological solution� Contents of the library� Statistical aspects� Demo� Prospects� Conclusions
Context: multiple uses.
GUI, ExcelAll, researchModelling / analysis
GUI, Excel, batch / SQL Server
Quarterly National Accounts, Balance of paiments
Temporal disaggregation
GUI, Excel, batch / SQL Server, Access
Quarterly National Accounts, Business surveys
Seasonal adjustment
GUI, batch / SQL Server
Balance of paimentsMissing values estimation/ forecasts
GUI, Excel, batch / DB2, SQL Server
Balance of paiments,National Accounts, External trade, Publications
Outliers detection
IT-EnvironmentProductTask
GOAL: Developing a tool that can be customized for each case.
Technological solution
� OO-Components• Managing the complexity• Reuse and extensibility
� Standard technologies• .NET (Windows), [COM]• Java
�→ NOT a closed application, but a set of modules / packages
Design
In-housedevelopments
(C#, VB.NET, Java, ...)
Nbb Libraries.NETJava
Commercial software
(Excel, ...)
Other [commercial] components
DBs.
Contents of the library (I)� Time series domain (specialized topics)
� Concepts• ARIMA models, TS regression, UCARIMA models (TRAMO-SEATS).• Structural models, state space forms (Durbin/Koopman).
� Algorithms• Kalman filters / smoothers (TRAMO, DK ).• Canonical decomposition, Wiener-Kolmogorov filters (SEATS).
• X11, ...
� Analysis tools• Residuals, ...
� Math / Stats domain� Concepts / algorithms related to time series problems.
• Polynomials (unit roots), linear filters• Matrix computation• Optimization• Statistical distributions, tests
Contents of the library (II)
� Tramo-Seats� = Most advanced application of the time series library.
� Completely new implementation: NOT a translation.
� Results as similar as possible of the FORTRAN ones.
� Open solution (extensions, GUI, Excel add-ins, ...)
� Easy use (no file-based I/O, just a few lines of code)
� Missing features• Fixed parameters (ARIMA model)• skipping approach for missings values, rarely used parameters
� New features• New outliers (SO, ...)• Kalman smoother estimation (SEATS)
Statistical aspects (I)
� Wiener-Kolmogorov filters� UCARIMA models (SEATS) extended to other models
(structural models, ...)
� SSF and Kalman filters / smoothers. � Used in:
• TRAMO.
• Many other applications.
� Based on:• TRAMO:
• Fast Chandrasekhar recursions.• GLS estimation (KF + QR decomposition), ...
• the Durbin-Koopman approach:• Diffuse initialization.• (Disturbance) smoother, ...
Statistical aspects (II) Seasonal adjustment
"Extended" SEATS(Burman-Wilson)
SEATS(Burman-Wilson)
WK filters
Durbin/Koopman+ Proietti
Durbin/KoopmanKalman smoother
Structural models (various seasonalcomponents)
UCARIMA models
+ X11 (based on the description of Ladiray/Quenneville)
DEMO
� Excel� User-defined functions (Tramo-Seats, X11, BSM).
� Add-In (outliers detection).
� Tramo-Seats (and others). Rich graphical interface.
� Just a few lines of code...
Demo (I). User-defined functions in Excel
Black-box use, designed for:
•fast analysis
•unskilled people
Demo (II). Excel Add-in: Color analyser
Outliers detection with TRAMO:
•Complete integration in a commercial software
•Use of the features of the hosting application
Demo (IV)Just a few lines of code
Steps:
•Creation of the series•Specifications (=input file)•Processing•Results retrieval:
•detailed information•dynamic computations
Prospects
� Library available at www.nbb.be/app/dqrd/index.htm (6/2006).
� Complete statistical framework (12/2006).� Extensions to other approaches
� Simulations� Band-pass methods� ...